Ear3 in Claude Code

Create, deploy, edit, and analyze AI voice interviews without leaving your editor. The Ear3 plugin adds two layers to Claude Code:

Skills — agent procedures that know how to do interview work (when to ask you, how to write a good prompt, what to report). They trigger from plain language: “create an interview about onboarding”.
MCP server — a set of typed ear3_* tools that do the work against the Ear3 API. Auth stays inside the server, so your secret key never appears in the conversation.

One install wires up both.

/plugin marketplace add https://www.ear3.ai/claude/marketplace.json
/plugin install ear3@ear3

Skills are the brain, the MCP server is the hands. The skill decides what should happen; the tools make it happen. You can use either alone, but together they give the smoothest result.

What you can do

Ask Claude…	Skill	Result
”Create an interview about X”	create-interview	Questions generated, deployed, a live shareable URL
”Change question 3” / “translate it”	edit-interview	Edited in place, or a dashboard link to edit by hand
”Add this interview to my app”	integrate-interview	The right SDK layer installed and wired
”Summarize the responses”	analyze-interview	A grounded findings report with verbatim quotes
”Export the responses to CSV”	export-responses	A CSV or Markdown file in your project

When to use which surface

Claude Code plugin (this page) — you work in an AI editor and want interviews created and wired for you from prompts.
CLI — the same power for a human at a terminal, or in scripts and CI.
SDK — hand-written integration when you want full control of the code.

All three hit the same /api/v1; the MCP server and CLI even share one login.

Other MCP clients

The MCP server is not Claude-Code-only. Any client that reads .mcp.json (Cursor, Windsurf, …) can use the ear3_* tools — see Setup & auth. The skills are Claude Code / agent procedures and can be copied into other agents’ skill folders.

Next: Install & authenticate.

Install & authenticate